Probably Approximately Correct
- The probably approximately correct (PAC)
learning model defines a setting and gives answers to our
questions in that setting.
- Leslie Valiant introduced PAC learning in
A theory of
the learnable. CACM 1984.
- Roughly, it tells us how many examples (and computation)
we will need to see before we can learn a hypothesis is
probably H, where H is approximately correct.
- PAC learning results are independent of the learning
algorithm used!